Offer complete three-rule table drafts with visible review guidance

docs/MAINTAINERS.md

@@ -313,6 +313,28 @@ regions suppress header hints. The same source region is used by document maps
 and fresh proposal verification. Ordinary Markdown and conversion gaps do not
 change, and matching source/region tokens do not excuse obsolete assignments.
 
+With `emit_table_proposals: true`, qualifying three-rule regions can also carry
+`review_candidate`: a proposed arrangement with `status: needs_review`, exact
+source-item assignments and line fragments, and inert HTML data cells. The
+same preview and review notice appear in the text response. It groups the top
+band provisionally and derives columns from whitespace corridors recurring
+across every nonoverlapping body line. All region items must be assigned once,
+every cell must contain source text, and the complete maximal partition must
+recur on every body line. Ragged, wrapped, straddling or over-budget layouts
+retain the evidence packet without a draft. Drafts are capped at 100 rows,
+12 columns, 400 source items, 20,000 HTML bytes and 40,000 canonical JSON bytes.
+
+Rows, columns and header roles remain unverified; no header tags, spans, blank
+values or new source text are inferred. Candidate digests bind algorithm,
+PDF SHA, region, assignments, fragments, uncertainty and presentation. When
+layout is supplied, semantic validation reconstructs the entire packet from
+it. Standalone output validation reconstructs the draft from its retained text
+and geometry as well, but cannot authenticate that packet to the original PDF;
+it does not replace fresh source replay. Document maps carry no candidate cells.
+Ordinary Markdown, its gap list and strict `verify_table_proposal` acceptance
+remain unchanged. Review the draft beside the source PDF before using its
+grouping; a visible draft is not an accepted table or verified extraction.
+
 Filesystem operations and rasterization happen locally, and PDF Tools does not
 upload files to a separate PDF service. Text, images, and metadata returned
 through MCP may still be processed under the selected host or model provider's

pdf-toolkit-mcp-share/server/index.js

@@ -3852,7 +3852,7 @@ async function listTools(request) {
             include_page_boundaries: { type: "boolean", description: "Include deterministic HTML comments marking page boundaries. Default: true." },
             compact: { type: "boolean", description: "Opt into counted dot-leader, isolated page-number, and Unicode-letter spaced-hyphen normalizations. Default: false." },
             remove_page_furniture: { type: "boolean", description: "Remove only source-evidenced extreme-margin page furniture and report every removed line by typed gap and counts. Set false to preserve all source lines. Default: true." },
-            emit_table_proposals: { type: "boolean", description: "Opt into emitting one bounded, deterministic table_proposals packet per abandoned table region (TABLE_TOPOLOGY_UNKNOWN or TABLE_RULING_UNSUPPORTED), carrying the region's text items, ruled and painted evidence, header hints, page, bbox, coordinate_space, and a source-and-IR-bound proposal_token so a host model can propose a structure for later read-only verification. Additive: the abstention gap is unchanged and default output stays byte-identical. Default: false." },
+            emit_table_proposals: { type: "boolean", description: "Opt into emitting one bounded, deterministic table_proposals packet per abandoned table region (TABLE_TOPOLOGY_UNKNOWN or TABLE_RULING_UNSUPPORTED), carrying the region's text items, ruled and painted evidence, header hints, page, bbox, coordinate_space, and a source-and-IR-bound proposal_token so a host model can propose a structure for later read-only verification. A narrowly qualifying captioned three-rule table may also include a draft arrangement and header-neutral HTML preview marked needs_review; its row/column grouping and header roles are unverified. Additive: the abstention gap is unchanged and default output stays byte-identical. Default: false." },
             allow_headerless_tables: { type: "boolean", description: "Opt into header-neutral HTML data cells for a complete source-covered ruled-rectangle cell grid without header evidence. No header roles, spans or blank data values are inferred; incomplete, ambiguous and cross-cell geometry remains text with typed gaps. Uses renderer profile 1.23.0; default false keeps profile 1.21.0 and unchanged output. No OCR or model." },
             output_path: { type: "string", description: "Optional absolute .md path, or ~/ path. The file is written only after complete bytes are staged and verified." },
             overwrite: { type: "boolean", description: "Replace an existing output_path only when its exact expected_output_identity is also supplied. Default: false." },
@@ -6082,6 +6082,7 @@ async function handleToolCall(request) {
             saved_output: savedOutput,
             ...(tableProposals !== undefined ? { table_proposals: tableProposals } : {}),
           };
+          const reviewCandidates = tableProposals?.filter(region => region.review_candidate) ?? [];
           const summary = [
             `Converted pages ${layout.page_range.start_page}-${layout.page_range.end_page} of ${fileName} to deterministic Markdown.`,
             `Status: ${rendered.conversion_status}. UTF-8 bytes: ${rendered.markdown_bytes}.`,
@@ -6091,8 +6092,14 @@ async function handleToolCall(request) {
             ...(pagesNeedingVision.length > 0
               ? [`Vision routing: use render_pdf_page for pages ${pagesNeedingVision.map(entry => entry.page).join(", ")}.`]
               : []),
-            "No OCR, external model, hidden link-target recovery, or table inference beyond complete source-bound text, ruled, or painted grids was performed.",
+            reviewCandidates.length > 0
+              ? "The optional table drafts use whitespace alignment and need review. No OCR, external model or hidden link-target recovery was performed."
+              : "No OCR, external model, hidden link-target recovery, or table inference beyond complete source-bound text, ruled, or painted grids was performed.",
             rendered.markdown,
+            ...reviewCandidates.map(region => [
+              `Page ${region.page}, ${region.region_id}: ${region.review_candidate.review_notice}`,
+              region.review_candidate.html,
+            ].join("\n\n")),
           ].join("\n\n");
           return {
             content: [{ type: "text", text: summary }],